Professional Documents
Culture Documents
17
2nd International Conference on Machine Learning and Computer Science(IMLCS'2013) August 25-26, 2013 Kuala Lumpur (Malaysia)
II. RELATED WORKS recognition to eliminate false alerts and motion levels are not
The rise in popularity of surveillance systems has been categorised. Mahalakshmi et al. [2] proposed an architecture
fuelled by the availability of cheap cameras and the ever for the setting up of a visual surveillance system using CCTV
increasing processing capacity of CPUs. Indeed, significant cameras. They also provided considerable technical details
research work has been done in this field during the past two about the equipment required and how to install and to use
decades. The following is a brief review of some previous them. Unfortunately, the system has not yet been tested and
works about visual surveillance systems. Yuan et al. [7] therefore results have yet to be provided. Rani [14] used an
proposed a distributed visual surveillance system by active infrared motion detector to detect motion in a home
connecting the output from multiple cameras to a local area environment. If the path between the radiation source and the
network. They used a simple background subtraction method sensor is interrupted, motion is assumed and this is linked to
to detect pedestrians and vehicles. However, their system the activity of an intruder.
could not recognise vehicles or pedestrians, it could only
detect their presence. The system was tested under controlled III. DESIGN AND IMPLEMENTATION
conditions which gave a detection accuracy of more than 95%.
However, the authors conceded that the system will perform
poorly under when it is raining, when there is little contrast
between the objects of interest and the background or when
there are many vehicles moving close to each other.
Lee et al. [8] developed a real-time threat assessment
system which could effectively distinguish between people,
animals and vehicles but cannot recognise them. They used
the contours of shapes obtained from each frame and
compared it with that in their database in order to differentiate
between people, animals and vehicles. An interesting work
was done by See and Lee [9] in which they proposed to have a
multi-modal home security system. They used a face
recognition module to authenticate the user and a motion
detection module to detect movement. However, the outputs
from these two systems were evaluated separately and were
not integrated into a single system to detect intruders.
Ansari et al. [10] described and implemented a sophisticated
motion detection system using Matlab to detect motion from a
video stream. The video is stored only when motion is
detected. However, they do not have different levels of motion
and owners are not given any warning messages. Nebel et al.
[11] reviewed some of the state-of-the-art approaches for
recognising human action from surveillance videos. They
showed that even the best performing systems could only
recognise a few distinct and basic actions with an accuracy of
only 70% under controlled conditions. Thus, they argue that
computer vision systems have not matured enough to be useful
in human action recognition. Kosba et al. [12] designed and
implemented a passive motion detection system which uses
the signal strength and waveform of wi-fi signals to detect
motion instead of relying on visual systems. They argue that
their system is better as it is not limited to line-of-sight vision Fig. 1 Algorithm to detect motion and recognise intruder
and it works even in the presence of smoke and at night. Their
current system gives a detection accuracy of 93%. However, Fig. 1 shows how the system operates. In the monitoring
changes in the physical layout of the building and interference mode, the camera monitors a specified area from an angle at a
from other devices can have an impact of their accuracy. They resolution of 640 x 480. In motion level exceeds a threshold
are also investigating the effect of size, shape and motion value of 15%, this triggers the face detection module. The
pattern on the performance of their system. motion level is calculated by taking the difference between the
Tuscano et al. [13] implemented a simple but effective current frame and the previous frame. If a person is detected,
surveillance system which upon detection of motion sounds an the face recognition algorithm tries to track and recognise the
alarm, start video recording, sends an email and send an MMS face by comparing it to the database of stored images. The
to the mobile phone of the security officer in charge of the face recognition module uses the Eigen face recognition
building. However, the system does not perform face algorithm to recognise and track faces. If that persons face is
18
2nd International Conference on Machine Learning and Computer Science(IMLCS'2013) August 25-26, 2013 Kuala Lumpur (Malaysia)
not present in the database, an alert is given which is then sent the frames and sends them to the email address specified by
to the owner mailbox. the owner of the house.
Visual surveillance systems have many advantages. Firstly,
IV. EVALUATION OF RESULTS they are not obstructive. Secondly, authentication and tracking
The system was tested in different light conditions. Motion can be done automatically from a distance. Thirdly, the setup
capture in normal daylight conditions is more responsive than requires simple low-cost cameras and a computer. Sensor-
night motion capture. This limitation is due to the fact that the based systems, in contrast, usually rely on some sort of contact
Infra-Red (IR) camera captures only one frame per second in a to detect movement. Furthermore, many surveillance
dark room while the in daylight it captures thirty frames per applications depend on one feature only whereas this system
second. Nevertheless, the system is able to detect motion uses two features which make it even more robust, reliable and
under both conditions. Thirteen different persons were asked effective. The proposed system can easily be implemented by
to walk in front of the camera, one at a time. All of them were anyone wishing to protect their home and its assets.
detected. However, the recognition rate was between only
62%. Furthermore, recognition works only up to a distance of V. CONCLUSION
7m. For detection, the moving object/person can be up 12m. The aim of this research was to develop an integrated
The system is designed to capture 30 frames per second (fps). intruder system by combining motion detection with face
If fewer frames are captured per second, some part of the recognition. The motion detector uses background subtraction
action in the event may be missed. 30fps is sufficient and to estimate the motion and then associates it with a certain risk
allow selection of useful frames to be retrieved and analysed. level. The face recognition distinguishes between authorized
These frames are also used to build the video, for streaming persons and intruders in order to eliminate false alerts. The
purposes. A emailing system was tested when the owner is at a system has been tested in real-life scenarios and this low-cost
remote place and he receives an intruder alert on his mobile home surveillance system shows good performance both
phone via the internet. However, a fast internet connection of during the day and at night when infra-red cameras were used.
at least 256 Mbps is required to have a quick response. For the The owner receives an email with the captured frame as
system to operate in real-time, images are at a resolution of attachment when a high risk alert is raised. The system can be
640 x 480 pixels. This allows for fast processing at 30fps. improved by adding more cameras so that one is present in
Small file size also allows for faster uploads and downloads. each room. However, additional work is required for the face
recognition module to increase its recognition ratio and to
make it less sensitive to variations in light intensity. .
REFERENCES
[1] Lockpick Library. 2011. Security Camera System. [ONLINE] Available
at:http://www.lockpicklibrary.com/cameras. [Accessed 2 May 13].
[2] Mahalakshmi R., Manjula M., Saranya S., Vaishnavi P., Shalini J. and
Arasa Kumar R. (2013). Intrusion Detection by Facial Recognition using
CCTV Cameras with Video Management System. International Journal
of Advanced Electrical and Electronics Engineering. 2(2), pp. 59-62.
[Accessed 2 May 13]
[3] Brodsky T., Cohen R., Cohen-Solal E., Gutta S., Lyons D., Philomin V.,
and Trajkovic M.: Visual Surveillance in retail stores and in the home,
in: Advanced Video-based Surveillance Systems (Kluwer Academic
Publishers, Boston, 2001), Chapter 4, pp. 50-61. [Accessed 2 May 13]
[4] Richard Szeliski, 2011. Computer Vision: Algorithms and Applications
(Texts in Computer Science). 1st ed. -: Springer.
[5] The Data Protection Act 2004 . [ONLINE] Available
at:http://www.gov.mu/portal/sites/ncbnew/files/DPA.pdf. [Accessed 24
May 2013].
[6] The Powers of Commissioner. [ONLINE] Available at:http://
dataprotection.gov.mu/English/AboutUs/The%20Commissioner/Pages/T
he-Powers-of-the-Commissioner-.aspx. [Accessed 24 May 2013].
[7] X. Yuan , Z. Sun, Y. Varol and G. Bebis (2003). A distributed Visual
Fig. 2 Difference images showing motion Surveillance System, IEEE International Conference on Advanced
Video and Signal Based Surveillance, Miami, July 2003.
Fig. 2 shows the instant when a person enters the field of [8] D.J. Lee, P. Zhan, A. Thomas, R. Schoenberger (2004). Shape-based
view and the system start to detect him from the starting point, Human Intrusion Detection, SPIE International Symposium on Defense
and Security, Visual Information Processing XIII, vol. 5438, pp. 81-91,
i.e., from a distance of 12 metres. This can be seen as some Orlando, Florida, USA, April 12-16, 2004.
dots from the first frame, indicating motion, then as the person [9] John See, Sze-Wei Lee (2007). An integrated vision-based architecture
come closer to the camera, the persons figure and outlines for home security system. IEEE Trans. Consumer Electronics, 53(2), pp.
489-498.
becomes more prominent. When the person comes within 7m, [10] Asif Ansari, T.C.Manjunath, C.Andril (2008) Implementation of a
the face recognition module will try to recognise the moving Motion Detection System, IEEE International Journal of Electrical and
object. If the face is recognised, no action is necessary. Computer Computer Engineering, 3(1), pp. 52-63.
However, if the face is not recognised the system will capture
19
2nd International Conference on Machine Learning and Computer Science(IMLCS'2013) August 25-26, 2013 Kuala Lumpur (Malaysia)
20